Temporal Back-Propagation Algorithm

1
Temporal Back-Propagation Algorithm

• Andrew J. Low

2
Problem Statement

• Program a Back-Propagation Algorithm to train a
  Multilayer Perceptron
• Programming Language C
• Distributed Time Lagged Feedforward Network

3
Distributed vs. Focused Networks

• Focused Network
• Additional Neurons added at input for memory
• Suitable for shift invariant series
• Distributed Network
• Memory added to each neuron in the form of
  additional inputs
• Allows for time variant series

4
Algorithm Detail

• Output Layer

• Hidden Layer

5
Results

• Program is incomplete at this time
• Advantages
• Broader range of problems can be solved
• Disadvantages
• Requires more processing power and memory
• Functionality
• Read Write Config Files, Config options
  including layers, memory depth, learning rate
• CLI Interface (Windows and Possibly Unix)
• Testing Series